Sparrows are one of the most costly pests for farmers, as they can reduce rice yields by 50-60%. Traditional control methods, such as the use of scarecrows, windmills, and pesticides, are often ineffective or cause negative impacts on the environment, such as damage to ecosystems and human health. To overcome this problem, YOLOv8-based object detection technology offers a modern solution to automatically detect bird pests with a high level of accuracy. This research aims to implement the YOLOv8 model on power-efficient embedded devices, such as Raspberry Pi 4 and Google Coral USB TPU Accelerator, to support real-time sparrow detection at an affordable cost. The research was conducted through three main stages, namely collecting bird image datasets to support model training, training the YOLOv8n model to produce reliable bird pest detection, and implementing the model on embedded devices with and without TPU accelerators to evaluate detection performance. The evaluation results show that the YOLOv8 model has high performance with precision 0.91, recall 0.86, mAP50 0.92, and mAP50-95 0.59 after being trained for 300 epochs. Implementation on Raspberry Pi 4 without accelerator only resulted in an inference speed of 0.39 Frame Per Second, while with Google Coral USB TPU, the speed increased significantly to 7 Frame Per Second. This proves that TPU accelerators are highly effective in supporting real-time object detection. This technology is expected to help farmers protect crops efficiently, reduce losses due to pests, support sustainable agricultural productivity, and contribute to the overall improvement of food security.